xref: /freebsd/sys/vm/vm_meter.c (revision 74bf4e164ba5851606a27d4feff27717452583e5)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)vm_meter.c	8.4 (Berkeley) 1/4/94
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/proc.h>
41 #include <sys/resource.h>
42 #include <sys/sx.h>
43 #include <sys/vmmeter.h>
44 #include <sys/smp.h>
45 
46 #include <vm/vm.h>
47 #include <vm/vm_page.h>
48 #include <vm/vm_extern.h>
49 #include <vm/vm_param.h>
50 #include <vm/pmap.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_object.h>
53 #include <sys/sysctl.h>
54 
55 /*
56  * Virtual memory MPSAFE temporary workarounds.
57  */
58 int debug_mpsafevm;
59 TUNABLE_INT("debug.mpsafevm", &debug_mpsafevm);
60 SYSCTL_INT(_debug, OID_AUTO, mpsafevm, CTLFLAG_RD, &debug_mpsafevm, 0,
61     "Enable/disable MPSAFE virtual memory support");
62 
63 struct vmmeter cnt;
64 
65 int maxslp = MAXSLP;
66 
67 SYSCTL_UINT(_vm, VM_V_FREE_MIN, v_free_min,
68 	CTLFLAG_RW, &cnt.v_free_min, 0, "");
69 SYSCTL_UINT(_vm, VM_V_FREE_TARGET, v_free_target,
70 	CTLFLAG_RW, &cnt.v_free_target, 0, "");
71 SYSCTL_UINT(_vm, VM_V_FREE_RESERVED, v_free_reserved,
72 	CTLFLAG_RW, &cnt.v_free_reserved, 0, "");
73 SYSCTL_UINT(_vm, VM_V_INACTIVE_TARGET, v_inactive_target,
74 	CTLFLAG_RW, &cnt.v_inactive_target, 0, "");
75 SYSCTL_UINT(_vm, VM_V_CACHE_MIN, v_cache_min,
76 	CTLFLAG_RW, &cnt.v_cache_min, 0, "");
77 SYSCTL_UINT(_vm, VM_V_CACHE_MAX, v_cache_max,
78 	CTLFLAG_RW, &cnt.v_cache_max, 0, "");
79 SYSCTL_UINT(_vm, VM_V_PAGEOUT_FREE_MIN, v_pageout_free_min,
80 	CTLFLAG_RW, &cnt.v_pageout_free_min, 0, "");
81 SYSCTL_UINT(_vm, OID_AUTO, v_free_severe,
82 	CTLFLAG_RW, &cnt.v_free_severe, 0, "");
83 
84 SYSCTL_STRUCT(_vm, VM_LOADAVG, loadavg, CTLFLAG_RD,
85     &averunnable, loadavg, "Machine loadaverage history");
86 
87 static int
88 vmtotal(SYSCTL_HANDLER_ARGS)
89 {
90 /* XXXKSE almost completely broken */
91 	struct proc *p;
92 	struct vmtotal total, *totalp;
93 	vm_map_entry_t entry;
94 	vm_object_t object;
95 	vm_map_t map;
96 	int paging;
97 	struct thread *td;
98 
99 	totalp = &total;
100 	bzero(totalp, sizeof *totalp);
101 	/*
102 	 * Mark all objects as inactive.
103 	 */
104 	GIANT_REQUIRED;
105 	mtx_lock(&vm_object_list_mtx);
106 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
107 		if (!VM_OBJECT_TRYLOCK(object)) {
108 			/*
109 			 * Avoid a lock-order reversal.  Consequently,
110 			 * the reported number of active pages may be
111 			 * greater than the actual number.
112 			 */
113 			continue;
114 		}
115 		vm_object_clear_flag(object, OBJ_ACTIVE);
116 		VM_OBJECT_UNLOCK(object);
117 	}
118 	mtx_unlock(&vm_object_list_mtx);
119 	/*
120 	 * Calculate process statistics.
121 	 */
122 	sx_slock(&allproc_lock);
123 	FOREACH_PROC_IN_SYSTEM(p) {
124 		if (p->p_flag & P_SYSTEM)
125 			continue;
126 		mtx_lock_spin(&sched_lock);
127 		switch (p->p_state) {
128 		case PRS_NEW:
129 			mtx_unlock_spin(&sched_lock);
130 			continue;
131 			break;
132 		default:
133 			FOREACH_THREAD_IN_PROC(p, td) {
134 				/* Need new statistics  XXX */
135 				switch (td->td_state) {
136 				case TDS_INHIBITED:
137 					if (TD_ON_LOCK(td) ||
138 					    (td->td_inhibitors ==
139 					    TDI_SWAPPED)) {
140 						totalp->t_sw++;
141 					} else if (TD_IS_SLEEPING(td) ||
142 					   TD_AWAITING_INTR(td) ||
143 					   TD_IS_SUSPENDED(td)) {
144 						if (td->td_priority <= PZERO)
145 							totalp->t_dw++;
146 						else
147 							totalp->t_sl++;
148 					}
149 					break;
150 
151 				case TDS_CAN_RUN:
152 					totalp->t_sw++;
153 					break;
154 				case TDS_RUNQ:
155 				case TDS_RUNNING:
156 					totalp->t_rq++;
157 					continue;
158 				default:
159 					break;
160 				}
161 			}
162 		}
163 		mtx_unlock_spin(&sched_lock);
164 		/*
165 		 * Note active objects.
166 		 */
167 		paging = 0;
168 		map = &p->p_vmspace->vm_map;
169 		vm_map_lock_read(map);
170 		for (entry = map->header.next;
171 		    entry != &map->header; entry = entry->next) {
172 			if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) ||
173 			    (object = entry->object.vm_object) == NULL)
174 				continue;
175 			VM_OBJECT_LOCK(object);
176 			vm_object_set_flag(object, OBJ_ACTIVE);
177 			paging |= object->paging_in_progress;
178 			VM_OBJECT_UNLOCK(object);
179 		}
180 		vm_map_unlock_read(map);
181 		if (paging)
182 			totalp->t_pw++;
183 	}
184 	sx_sunlock(&allproc_lock);
185 	/*
186 	 * Calculate object memory usage statistics.
187 	 */
188 	mtx_lock(&vm_object_list_mtx);
189 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
190 		/*
191 		 * Perform unsynchronized reads on the object to avoid
192 		 * a lock-order reversal.  In this case, the lack of
193 		 * synchronization should not impair the accuracy of
194 		 * the reported statistics.
195 		 */
196 		if (object->type == OBJT_DEVICE) {
197 			/*
198 			 * Devices, like /dev/mem, will badly skew our totals.
199 			 */
200 			continue;
201 		}
202 		totalp->t_vm += object->size;
203 		totalp->t_rm += object->resident_page_count;
204 		if (object->flags & OBJ_ACTIVE) {
205 			totalp->t_avm += object->size;
206 			totalp->t_arm += object->resident_page_count;
207 		}
208 		if (object->shadow_count > 1) {
209 			/* shared object */
210 			totalp->t_vmshr += object->size;
211 			totalp->t_rmshr += object->resident_page_count;
212 			if (object->flags & OBJ_ACTIVE) {
213 				totalp->t_avmshr += object->size;
214 				totalp->t_armshr += object->resident_page_count;
215 			}
216 		}
217 	}
218 	mtx_unlock(&vm_object_list_mtx);
219 	totalp->t_free = cnt.v_free_count + cnt.v_cache_count;
220 	return (sysctl_handle_opaque(oidp, totalp, sizeof total, req));
221 }
222 
223 /*
224  * vcnt() -	accumulate statistics from all cpus and the global cnt
225  *		structure.
226  *
227  *	The vmmeter structure is now per-cpu as well as global.  Those
228  *	statistics which can be kept on a per-cpu basis (to avoid cache
229  *	stalls between cpus) can be moved to the per-cpu vmmeter.  Remaining
230  *	statistics, such as v_free_reserved, are left in the global
231  *	structure.
232  *
233  * (sysctl_oid *oidp, void *arg1, int arg2, struct sysctl_req *req)
234  */
235 static int
236 vcnt(SYSCTL_HANDLER_ARGS)
237 {
238         int error = 0;
239 	int count = *(int *)arg1;
240 #ifdef SMP
241 	int i;
242 	int offset = (char *)arg1 - (char *)&cnt;
243 
244 	for (i = 0; i < mp_ncpus; ++i) {
245 		struct pcpu *pcpu = pcpu_find(i);
246 		count += *(int *)((char *)&pcpu->pc_cnt + offset);
247 	}
248 #else
249 	int offset = (char *)arg1 - (char *)&cnt;
250 	count += *(int *)((char *)PCPU_PTR(cnt) + offset);
251 #endif
252 	error = SYSCTL_OUT(req, &count, sizeof(int));
253 	return(error);
254 }
255 
256 SYSCTL_PROC(_vm, VM_TOTAL, vmtotal, CTLTYPE_OPAQUE|CTLFLAG_RD,
257     0, sizeof(struct vmtotal), vmtotal, "S,vmtotal",
258     "System virtual memory statistics");
259 SYSCTL_NODE(_vm, OID_AUTO, stats, CTLFLAG_RW, 0, "VM meter stats");
260 SYSCTL_NODE(_vm_stats, OID_AUTO, sys, CTLFLAG_RW, 0, "VM meter sys stats");
261 SYSCTL_NODE(_vm_stats, OID_AUTO, vm, CTLFLAG_RW, 0, "VM meter vm stats");
262 SYSCTL_NODE(_vm_stats, OID_AUTO, misc, CTLFLAG_RW, 0, "VM meter misc stats");
263 
264 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_swtch, CTLTYPE_UINT|CTLFLAG_RD,
265 	&cnt.v_swtch, 0, vcnt, "IU", "Context switches");
266 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_trap, CTLTYPE_UINT|CTLFLAG_RD,
267 	&cnt.v_trap, 0, vcnt, "IU", "Traps");
268 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_syscall, CTLTYPE_UINT|CTLFLAG_RD,
269 	&cnt.v_syscall, 0, vcnt, "IU", "Syscalls");
270 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_intr, CTLTYPE_UINT|CTLFLAG_RD,
271 	&cnt.v_intr, 0, vcnt, "IU", "Hardware interrupts");
272 SYSCTL_PROC(_vm_stats_sys, OID_AUTO, v_soft, CTLTYPE_UINT|CTLFLAG_RD,
273 	&cnt.v_soft, 0, vcnt, "IU", "Software interrupts");
274 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vm_faults, CTLTYPE_UINT|CTLFLAG_RD,
275 	&cnt.v_vm_faults, 0, vcnt, "IU", "VM faults");
276 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_cow_faults, CTLTYPE_UINT|CTLFLAG_RD,
277 	&cnt.v_cow_faults, 0, vcnt, "IU", "COW faults");
278 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_cow_optim, CTLTYPE_UINT|CTLFLAG_RD,
279 	&cnt.v_cow_optim, 0, vcnt, "IU", "Optimized COW faults");
280 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_zfod, CTLTYPE_UINT|CTLFLAG_RD,
281 	&cnt.v_zfod, 0, vcnt, "IU", "Zero fill");
282 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_ozfod, CTLTYPE_UINT|CTLFLAG_RD,
283 	&cnt.v_ozfod, 0, vcnt, "IU", "Optimized zero fill");
284 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swapin, CTLTYPE_UINT|CTLFLAG_RD,
285 	&cnt.v_swapin, 0, vcnt, "IU", "Swapin operations");
286 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swapout, CTLTYPE_UINT|CTLFLAG_RD,
287 	&cnt.v_swapout, 0, vcnt, "IU", "Swapout operations");
288 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swappgsin, CTLTYPE_UINT|CTLFLAG_RD,
289 	&cnt.v_swappgsin, 0, vcnt, "IU", "Swapin pages");
290 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_swappgsout, CTLTYPE_UINT|CTLFLAG_RD,
291 	&cnt.v_swappgsout, 0, vcnt, "IU", "Swapout pages");
292 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodein, CTLTYPE_UINT|CTLFLAG_RD,
293 	&cnt.v_vnodein, 0, vcnt, "IU", "Vnodein operations");
294 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodeout, CTLTYPE_UINT|CTLFLAG_RD,
295 	&cnt.v_vnodeout, 0, vcnt, "IU", "Vnodeout operations");
296 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodepgsin, CTLTYPE_UINT|CTLFLAG_RD,
297 	&cnt.v_vnodepgsin, 0, vcnt, "IU", "Vnodein pages");
298 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vnodepgsout, CTLTYPE_UINT|CTLFLAG_RD,
299 	&cnt.v_vnodepgsout, 0, vcnt, "IU", "Vnodeout pages");
300 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_intrans, CTLTYPE_UINT|CTLFLAG_RD,
301 	&cnt.v_intrans, 0, vcnt, "IU", "In transit page blocking");
302 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_reactivated, CTLTYPE_UINT|CTLFLAG_RD,
303 	&cnt.v_reactivated, 0, vcnt, "IU", "Reactivated pages");
304 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_pdwakeups, CTLTYPE_UINT|CTLFLAG_RD,
305 	&cnt.v_pdwakeups, 0, vcnt, "IU", "Pagedaemon wakeups");
306 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_pdpages, CTLTYPE_UINT|CTLFLAG_RD,
307 	&cnt.v_pdpages, 0, vcnt, "IU", "Pagedaemon page scans");
308 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_dfree, CTLTYPE_UINT|CTLFLAG_RD,
309 	&cnt.v_dfree, 0, vcnt, "IU", "");
310 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_pfree, CTLTYPE_UINT|CTLFLAG_RD,
311 	&cnt.v_pfree, 0, vcnt, "IU", "");
312 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_tfree, CTLTYPE_UINT|CTLFLAG_RD,
313 	&cnt.v_tfree, 0, vcnt, "IU", "");
314 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_page_size, CTLTYPE_UINT|CTLFLAG_RD,
315 	&cnt.v_page_size, 0, vcnt, "IU", "");
316 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_page_count, CTLTYPE_UINT|CTLFLAG_RD,
317 	&cnt.v_page_count, 0, vcnt, "IU", "");
318 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_free_reserved, CTLTYPE_UINT|CTLFLAG_RD,
319 	&cnt.v_free_reserved, 0, vcnt, "IU", "");
320 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_free_target, CTLTYPE_UINT|CTLFLAG_RD,
321 	&cnt.v_free_target, 0, vcnt, "IU", "");
322 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_free_min, CTLTYPE_UINT|CTLFLAG_RD,
323 	&cnt.v_free_min, 0, vcnt, "IU", "");
324 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_free_count, CTLTYPE_UINT|CTLFLAG_RD,
325 	&cnt.v_free_count, 0, vcnt, "IU", "");
326 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_wire_count, CTLTYPE_UINT|CTLFLAG_RD,
327 	&cnt.v_wire_count, 0, vcnt, "IU", "");
328 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_active_count, CTLTYPE_UINT|CTLFLAG_RD,
329 	&cnt.v_active_count, 0, vcnt, "IU", "");
330 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_inactive_target, CTLTYPE_UINT|CTLFLAG_RD,
331 	&cnt.v_inactive_target, 0, vcnt, "IU", "");
332 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_inactive_count, CTLTYPE_UINT|CTLFLAG_RD,
333 	&cnt.v_inactive_count, 0, vcnt, "IU", "");
334 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_cache_count, CTLTYPE_UINT|CTLFLAG_RD,
335 	&cnt.v_cache_count, 0, vcnt, "IU", "");
336 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_cache_min, CTLTYPE_UINT|CTLFLAG_RD,
337 	&cnt.v_cache_min, 0, vcnt, "IU", "");
338 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_cache_max, CTLTYPE_UINT|CTLFLAG_RD,
339 	&cnt.v_cache_max, 0, vcnt, "IU", "");
340 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_pageout_free_min, CTLTYPE_UINT|CTLFLAG_RD,
341 	&cnt.v_pageout_free_min, 0, vcnt, "IU", "");
342 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_interrupt_free_min, CTLTYPE_UINT|CTLFLAG_RD,
343 	&cnt.v_interrupt_free_min, 0, vcnt, "IU", "");
344 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_forks, CTLTYPE_UINT|CTLFLAG_RD,
345 	&cnt.v_forks, 0, vcnt, "IU", "Number of fork() calls");
346 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vforks, CTLTYPE_UINT|CTLFLAG_RD,
347 	&cnt.v_vforks, 0, vcnt, "IU", "Number of vfork() calls");
348 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_rforks, CTLTYPE_UINT|CTLFLAG_RD,
349 	&cnt.v_rforks, 0, vcnt, "IU", "Number of rfork() calls");
350 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_kthreads, CTLTYPE_UINT|CTLFLAG_RD,
351 	&cnt.v_kthreads, 0, vcnt, "IU", "Number of fork() calls by kernel");
352 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_forkpages, CTLTYPE_UINT|CTLFLAG_RD,
353 	&cnt.v_forkpages, 0, vcnt, "IU", "VM pages affected by fork()");
354 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_vforkpages, CTLTYPE_UINT|CTLFLAG_RD,
355 	&cnt.v_vforkpages, 0, vcnt, "IU", "VM pages affected by vfork()");
356 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_rforkpages, CTLTYPE_UINT|CTLFLAG_RD,
357 	&cnt.v_rforkpages, 0, vcnt, "IU", "VM pages affected by rfork()");
358 SYSCTL_PROC(_vm_stats_vm, OID_AUTO, v_kthreadpages, CTLTYPE_UINT|CTLFLAG_RD,
359 	&cnt.v_kthreadpages, 0, vcnt, "IU", "VM pages affected by fork() by kernel");
360 
361 SYSCTL_INT(_vm_stats_misc, OID_AUTO,
362 	zero_page_count, CTLFLAG_RD, &vm_page_zero_count, 0, "");
363 #if 0
364 SYSCTL_INT(_vm_stats_misc, OID_AUTO,
365 	page_mask, CTLFLAG_RD, &page_mask, 0, "");
366 SYSCTL_INT(_vm_stats_misc, OID_AUTO,
367 	page_shift, CTLFLAG_RD, &page_shift, 0, "");
368 SYSCTL_INT(_vm_stats_misc, OID_AUTO,
369 	first_page, CTLFLAG_RD, &first_page, 0, "");
370 SYSCTL_INT(_vm_stats_misc, OID_AUTO,
371 	last_page, CTLFLAG_RD, &last_page, 0, "");
372 #endif
373